Three-dimensional dynamic monitoring of crevasses based on deep learning and surface elevation reconstruction methods

Abstract

The evolution of crevasses on ice shelves plays a pivotal role in maintaining their stability. Investigating the developmental pattern of crevasses on ice shelf surfaces necessitates long-term dynamic monitoring, high-precision automatic extraction algorithms, and quantitative analysis tools. Accordingly, based on characteristics such as the long span of crevasses and their meandering distribution, we proposed Strip_Unet for the identification of crevasses. Within this, the Strip_block module utilizes strip convolution in conjunction with traditional convolution to capture the contextual information from the surrounding area, and the ASPP module is used to enhance the receptive field. Leveraging this model, we successfully identified crevasses within the damaged area of the Amery Ice Shelf from 2003 through 2023, utilizing 15-meter resolution Landsat imagery. Subsequently, we retrieved the depth details of these crevasses by applying ATL06 data, allowing us to quantitatively assess the evolution of crevasses. Our study represents the first investigation into the development of the damage area on the Amery Ice Shelf through long-term time-series analysis. Our findings indicate that crevasses originate from specific areas and follow recurring propagation cycles, exhibiting a consistent annual increase in total length. Additionally, changes in the width of these crevasses coincide with alterations in their depth. Notably, the downstream regions of the damaged area demonstrate higher crevasses density and the fastest movement speed. The proposed automatic identification of crevasses and three-dimensional dynamic monitoring hold significant potential for application throughout the Antarctic region, providing valuable insights into the long-term stability of ice shelves.